GA-SVM based feature selection and parameters optimization for BCI research

Brain Computer Interface (BCI) can translate the mind of the patients who suffered from locked- in syndrome into control commands or meaning symbols. Using this technology, the patients can communicate with the world. The core parts of a typical BCI system is feature extraction and pattern recognition. Too many irrelevant and redundant features will increase the time of classification and decrease the prediction accuracy. The kernel parameters setting for support vector machine (SVM) also impact on the classification accuracy. In this paper, after the features extracted though the algorithm called Sample Entropy, GA-SVM hybrid algorithm was used with two purposes: Selecting of the optimal feature subset and deciding the parameters for SVM classifier. Compared with GA-based feature selection and GA-based parameters optimization for SVM, the GA-SVM hybrid algorithm has fewer input features and gain much higher classification accuracy.